Downloads · 30 days
9
5% of all-time downloads
pei-germany/biom-bert-smm4h2026-task5-patient-metadata
biom-bert-smm4h2026-task5-patient-metadata is a machine learning model from pei-germany. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
This model was developed for SMM4H-HeaRD 2026 Task 5, which focuses on detecting patient metadata in SARS-CoV-2 sequencing articles.
Downloads · 30 days
9
5% of all-time downloads
All-time downloads
168
Public
Parameters
335M
1.3 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors1.3 GB · 100%
From the Hugging Face model README
This model was developed for SMM4H-HeaRD 2026 Task 5, which focuses on detecting patient metadata in SARS-CoV-2 sequencing articles.
Patient metadata include demographic, clinical, and epidemiological information such as age, outcomes, hospitalization, and comorbidities.
Our system is based on BioM-BERT-PubMed-PMC-Large and leverages a Natural Language Inference (NLI) formulation combined with paraphrase-based data augmentation. The model achieves a higher F1 score (0.786) compared to both the mean (0.729) and median (0.754) performance across all participating systems.
Our results:
📄 See our system description paper: (https://aclanthology.org/2026.smm4h-1.24/)
👩💻 Authors:
Given a sentence, the model predicts:
| Model | Precision | Recall | F1 |
|---|---|---|---|
| BioM-BERT (ours) | 0.797 | 0.774 | 0.786 |
| Mean (all teams) | 0.712 | 0.756 | 0.729 |
| Median (all teams) | 0.741 | 0.772 | 0.754 |
👉 Our model achieves a higher F1 score than both the mean and median performance across all participating systems.
This model is formulated as a Natural Language Inference (NLI) task.
Each input sentence is paired with a fixed hypothesis, and the model predicts whether the sentence entails the hypothesis.
from transformers import AutoTokenizer, AutoModelForSequenceClassification
import torch
model_name = "pei-germany/biom-bert-smm4h2026-task5-patient-metadata"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSequenceClassification.from_pretrained(model_name)
model.eval()
text = "The patient was admitted to the ICU and required ventilation."
hypothesis = "This sentence explicitly describes patients, cases, or patient-related clinical or epidemiological information."
inputs = tokenizer(text, hypothesis, return_tensors="pt", truncation=True)
with torch.no_grad():
logits = model(**inputs).logits
probs = torch.softmax(logits, dim=-1)
entailment_prob = probs[:, -1]
prediction = int(entailment_prob > 0.5)
print("Prediction:", prediction)